Rank-based interolog mapping for predicting proteinprotein interactions between genomes

As rapidly increasing number of sequenced genomes, the methods for predicting protein-protein interactions (PPIs) from one organism to another is becoming important. Best-match and generalized interolog mapping methods have been proposed for predicting (PPIs). However, best-match mapping method suffers from low coverage of the total interactome, because of using only best matches. Generalized interolog mapping method may predict unreliable interologs at a certain similarity cutoff, because of the homologs differed in subcellular compartment, biological process, or function from the query protein. Here, we propose a new “rank-based interolog mapping” method, which uses the pairs of proteins with high sequence similarity (E-value<;10^-10) and ranked by BLASTP E-value in all possible homologs to predict interologs. First, we evaluated “rank-based interolog mapping” on predicting the PPIs in yeast. The accuracy at selecting top 5 and top 10 homologs are 0.17, and 0.12, respectively, and our method outperformed generalized interolog mapping method (accuracy=0.04) with the joint E-value<;10^-70. Furthermore, our method was used to predict PPIs in four organisms, including worm, fly, mouse, and human. In addition, we used Gene Ontology (GO) terms to analyzed PPIs, which reflect cellular component, biological process, and molecular function, inferred by rank-based mapping method. Our rank-based mapping method can predict more reliable interactions under a given percentage of false positives than the best-match and generalized interolog mapping methods. We believe that the rank-based mapping method is useful method for predicting PPIs in a genome-wide scale.